A method and control system for intelligent paving and compaction control in road construction

Abstract

This invention belongs to the field of road construction technology. It provides an intelligent paving and compaction control method and system for road construction, comprising: constructing a calculation model for joint compaction conditions; using a vision system and high-precision positioning to identify joint lines in real time, and combining this with the steel wheel trajectory to determine whether the roller has entered the joint compaction condition; if so, analyzing the symmetry of the vibration data of the left and right steel wheels to determine the degree of interference with the original CCC signal; if the degree of interference is high, activating the joint compaction condition calculation model, and combining current road construction data to filter out the old material support effect, outputting the equivalent stiffness of the new material, and comparing it with the target stiffness range to determine whether compaction control intervention is needed; if needed, using the equivalent stiffness of the new material as the control target, and combining the roller speed and amplitude constraints, predicting and outputting the optimal amplitude and speed adjustment sequence to ensure that the new material side of the longitudinal joint can achieve the same density and stiffness standards as the main road surface area.

Description

Technical Field

[0001] This invention belongs to the field of road construction technology, specifically an intelligent paving and compaction control method and control system for road construction. Background Technology

[0002] In the field of intelligent road construction, continuous compaction control (CCC) systems have become a key technology for improving the uniformity of compaction quality and operational efficiency. This system analyzes the dynamic response (such as vibration acceleration) of the interaction between the roller's steel wheel and the road surface in real time, calculates and feeds back the degree of material compaction (stiffness / modulus), and then automatically adjusts parameters such as vibration amplitude and frequency, forming a closed loop of "measurement-feedback-control". However, this technology faces a fundamental problem when dealing with the common and critical construction link of longitudinal joints: when the roller performs "joint compaction" (i.e., the steel wheel spans between already compacted, cold, hard old material and newly laid, hot, soft new material), the stiffness of the foundation under the steel wheel exhibits a drastically asymmetrical distribution. At this time, the signal collected by the CCC sensor is essentially a "mixed signal" resulting from the coupling of the physical properties of the old and new materials. Because the stiffness of the already compacted old material is much higher than that of the new material, its strong "support effect" significantly increases the overall feedback value, causing the control system to severely underestimate the actual compaction requirements of the new material area. If closed-loop control is directly implemented based on this interfered signal, the system will automatically reduce the compaction energy output, resulting in insufficient compaction of the new material side at the joint, creating a weak point in long-term durability. Existing general-purpose CCC systems lack the ability to automatically identify and decouple signals for this special working condition, still relying on manual intervention based on operator experience, which has become one of the main bottlenecks restricting uniform compaction of the entire road surface and intelligent management of construction quality.

[0003] Therefore, the present invention provides an intelligent paving and compaction control method and control system for road construction. Summary of the Invention

[0004] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0005] The technical solution adopted by this invention to solve its technical problem is: an intelligent paving and compaction control method for road construction, comprising the following steps:

[0006] Step S10: Train and construct a solution model for the rolling compaction condition by using historical road construction data;

[0007] Step S20: The joint line is identified in real time through the vision system and high-precision positioning, and the steel wheel trajectory is combined to determine whether the road roller has entered the joint compaction condition. If it has, the degree of interference of the original CCC signal is determined by analyzing the symmetry of the vibration data of the left and right steel wheels.

[0008] Step S30: If the degree of interference is high, the calculation model for the saddle-joint compaction condition is activated, and the old material support effect is filtered out by combining the current road construction data. The equivalent stiffness of the new material is output and compared with the target stiffness range to determine whether compaction control intervention is required.

[0009] Step S40: If necessary, take the equivalent stiffness of the new material as the control target, and combine the roller speed and amplitude constraints to predict and output the optimal amplitude and speed adjustment sequence.

[0010] As a further aspect of the present invention: historical road construction data includes: roller body sensor data, specifically the original CCC signal, high-frequency acceleration time-domain data of independent vibration of the left and right steel wheels, roller amplitude, frequency and travel speed;

[0011] Material temperature data, including the coverage ratio of new material area in the seam area, the average temperature of new material area, the average temperature of old material area, and the temperature gradient between new and old material areas;

[0012] The actual stiffness data, specifically the on-site test values ​​on the new material side of the joint, are obtained by point testing using a portable drop weight deflectometer and nuclear density meter after compaction along the joint.

[0013] As a further aspect of the present invention: the joint-spanning compaction condition specifically refers to the specific operational state in which the roller steel wheel's compaction trajectory laterally crosses the compacted and cooled old pavement and the newly paved hot and soft new material during the construction of longitudinal joints of asphalt pavement.

[0014] During the training and construction of the calculation model for the rolling joint compaction condition, gradient boosting decision tree or deep feedforward neural network is selected as the basic model architecture. The roller body sensor data and material temperature data are used as input features, and the real stiffness data is used as output features to integrate into a data feature group. All data in the data feature group are collected from the same new and old material area at the same time.

[0015] As a further aspect of the present invention: during the training and construction of the calculation model for the rolling and compaction condition, the data feature group is divided into a training set, a validation set, and a test set. The mean squared error is used as the loss function, and the model parameters are iteratively optimized on the training set using the backpropagation algorithm or the gradient boosting algorithm.

[0016] After each training round, the loss is calculated on the validation set. When the validation set loss no longer decreases for several consecutive training rounds, the training is terminated early and the model parameters with the lowest validation loss are rolled back.

[0017] Using the validation set, key hyperparameters such as learning rate, number of network layers and nodes, and depth and number of GBDT trees were fined through grid search or Bayesian optimization. Finally, the model was tested using the test set, and the model with the coefficient of determination closest to 1 was selected as the solution model for the saddle-compact rolling condition.

[0018] As a further aspect of the present invention, the specific operation for determining whether the road roller has entered the joint compaction condition is as follows:

[0019] Ground images are acquired through a vision system, and the location coordinates and attitude information of the road roller are obtained using the GNSS / INS system on the road roller itself.

[0020] The joint lines in the ground image are identified by the edge detection algorithm. Based on the camera calibration parameters and the real-time attitude of the roller, the 2D pixel coordinates in the ground image are converted into 3D local coordinates centered on the roller. Combined with the roller's position coordinates and attitude, the joint lines are mapped into 3D spatial curves in the global coordinate system, thus obtaining the joint line spatial curves.

[0021] Based on the position coordinates and attitude information of the road roller, the precise movement trajectory of the grounding center points of the left and right steel wheels is calculated using the dead reckoning method. The shortest lateral distance between the steel wheel trajectory and the joint line spatial curve is calculated. If the distance meets the requirements for entering the joint compaction condition, it is determined that the joint compaction condition has been entered; otherwise, it has not been entered.

[0022] As a further aspect of the present invention, the operation for determining the degree of interference of the original CCC signal is as follows:

[0023] After confirming that the rolling condition has been entered, the vibration signals of the left and right steel wheels of the roller are acquired, and the ratio of the effective values ​​of the vibration signals of the left and right steel wheels is calculated to obtain the amplitude symmetry ratio.

[0024] If the amplitude symmetry ratio is not within the amplitude symmetry range, it indicates that the original CCC signal is highly disturbed; conversely, it indicates a low degree of disturbance.

[0025] As a further aspect of the present invention: when outputting the equivalent stiffness of new material, the existing calculation model for the saddle-joint compaction condition is activated, and the current road construction data with the same data type as the historical road construction data is input. The old material support effect is filtered out through model calculation, and the equivalent stiffness of new material is directly output.

[0026] When determining whether compaction control intervention is needed, the equivalent stiffness of the new material is compared with the target stiffness range. The target stiffness range is constructed from the target stiffness and the allowable stiffness error during road construction. If the equivalent stiffness of the new material is within the target stiffness range, then compaction control intervention is not required.

[0027] If the equivalent stiffness of the new material is less than the minimum value of the target stiffness range, the support effect of the old material usually seriously underestimates the actual compaction requirements of the new material area. Therefore, the stiffness of the new material area is insufficient, and compaction control intervention is required.

[0028] As a further aspect of the present invention: the process of predicting and outputting the optimal amplitude and velocity adjustment sequence includes:

[0029] The optimization time window is divided into several time steps of equal length. Under amplitude and velocity constraints, a set of candidate amplitude-velocity operation pairs is randomly selected for each time step. All operation pairs are integrated in time sequence to obtain the candidate operation sequence.

[0030] The candidate operation sequence is input into the state prediction model, the predicted output is the equivalent stiffness of the new material, and the candidate operation sequence that meets the control target requirements is selected as the usable operation sequence, while the candidate operation sequence that does not meet the control target requirements is removed.

[0031] Second-order difference calculations are performed on each available operation sequence to obtain the smoothness characterization value of the available operation sequence. The available operation sequence with the smallest smoothness characterization value is selected as the optimal amplitude and velocity adjustment sequence.

[0032] As a further aspect of the present invention, the state prediction model is constructed as follows: the equivalent stiffness of new material, roller speed, current amplitude level, and material temperature gradient are extracted from historical road construction data as input features, the equivalent stiffness of new material is used as output feature, a deep feedforward neural network is used as the basic framework model, and the state prediction model is obtained by training and verification through historical road construction data.

[0033] The constraints include amplitude constraints and velocity constraints. Amplitude constraints include available discrete speeds, and velocity constraints include the allowable speed range and maximum acceleration and deceleration limits. At the same time, the optimization time window constraint is determined according to the road roller control system. The control objective is that when the optimization time window ends, the equivalent stiffness of the new material reaches the minimum target stiffness or is within the target stiffness range.

[0034] An intelligent paving and compaction control system for road construction, the system comprising:

[0035] Solution model construction module: The solution model for the rolling compaction condition with joints is constructed by training on historical road construction data;

[0036] Interference analysis and judgment module: The joint line is identified in real time through vision system and high-precision positioning, and the steel wheel trajectory is combined to determine whether the road roller has entered the joint compaction condition. If it has, the degree of interference of the original CCC signal is judged by analyzing the symmetry of the vibration data of the left and right steel wheels.

[0037] Compaction Intervention Judgment Module: If the degree of interference is high, the calculation model of the rolling condition with joints is activated, and the old material support effect is filtered out by combining the current road construction data. The equivalent stiffness of the new material is output and compared with the target stiffness range to determine whether compaction control intervention is required.

[0038] Compaction control output module: If necessary, the equivalent stiffness of the new material is used as the control target, and the optimal amplitude and speed adjustment sequence is predicted and output in combination with the speed and amplitude constraints of the roller.

[0039] The beneficial effects of this invention are as follows: First, by constructing a dedicated solution model through machine learning, the "equivalent stiffness of the new material," representing the true compaction state of the new material, is accurately separated from the disturbed mixed signals, thus solving the problem of measurement distortion. Second, by integrating vision and positioning technologies, automatic and accurate identification of joint compaction conditions is achieved, and online diagnosis of signal reliability is performed. Third, based on the calculated true stiffness and comparison with construction standards, the necessity of intervention is intelligently determined. Finally, through model predictive control methods, a set of specific, optimized, and smooth amplitude and speed adjustment instructions is automatically generated, changing the traditional joint compaction operation mode that relies on manual experience. The system can automatically identify joint compaction conditions, eliminate interference in real time, accurately determine insufficient new material compaction, and output a compaction control adjustment scheme. This ensures that the new material side of the longitudinal joint can achieve the same density and stiffness standards as the main pavement area, eliminating early pavement risks such as cracking and water seepage caused by uneven compaction. Attached Figure Description

[0040] The invention will now be further described with reference to the accompanying drawings.

[0041] Figure 1 This is a flowchart illustrating the steps of an intelligent paving and compaction control method for road construction as described in an embodiment of the present invention.

[0042] Figure 2 This is a logic judgment diagram of an intelligent paving and compaction control method for road construction as described in an embodiment of the present invention;

[0043] Figure 3 This is a flowchart of an intelligent paving and compaction control system for road construction as described in an embodiment of the present invention. Detailed Implementation

[0044] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0045] Example 1: Please refer to Figure 1-2 As shown in the embodiment of the present invention, an intelligent paving and compaction control method for road construction includes the following steps:

[0046] Step S10: Train and construct a solution model for the rolling compaction condition by using historical road construction data;

[0047] In step S10, the historical road construction data includes, but is not limited to, roller body sensor data, material temperature data, and actual stiffness data;

[0048] Among them, the sensor data of the road roller body includes the original CCC signal (stiffness value calculated by the system), independent vibration signals of the left and right steel wheels, namely the high-frequency acceleration time-domain data of the sensors on the left and right sides (used to calculate the asymmetry characteristics: cross-correlation coefficient), road roller amplitude, frequency, and travel speed.

[0049] The sensor data of the road roller body is exported from the on-board data recorder of the road roller.

[0050] Material temperature data includes the coverage ratio of the new material area in the seam area, the average temperature of the new material area, the average temperature of the old material area, and the temperature gradient between the new and old material areas;

[0051] Material temperature data is recorded by historically archived infrared devices and exported from the monitoring system;

[0052] The actual stiffness data includes the actual stiffness value of the new material side (the stiffness value obtained by point detection on the new material side of the joint immediately after compaction using equipment such as portable falling weight deflectometer (PFWD) and nuclear density meter).

[0053] The actual stiffness data is obtained through on-site testing of the new material side of the joint;

[0054] In step S10, the joint-spanning compaction condition specifically refers to the specific working state in the construction of longitudinal joints of asphalt pavement, where the compaction trajectory of the roller's steel wheel laterally crosses the boundary line between the compacted and cooled old pavement (or the cold joint material laid the day before) and the newly paved hot asphalt mixture.

[0055] In step S10, the process of training and constructing the solution model for the rolling compaction condition is as follows:

[0056] Model selection: Considering the complex spatiotemporal coupling and nonlinear relationships between input features (such as the nonlinear interaction between vibration signals and temperature and velocity), gradient boosting decision trees or deep feedforward neural networks are selected as the basic model architecture.

[0057] Data feature construction: The data feature group is obtained by integrating the sensor data of the roller body and the material temperature data as input features (such as cross-correlation coefficient, stiffness value, roller amplitude, frequency, travel speed, coverage ratio of new material area, average temperature of new material area, average temperature of old material area and temperature gradient of new and old material areas) and the real stiffness data as output features.

[0058] It should be noted that the data in the data feature group were all collected from the same new and old material area at the same time;

[0059] Example of a model structure (using a deep feedforward neural network as an example):

[0060] Input layer (number of nodes = feature dimension, e.g., 9-dimensional);

[0061] Hidden layer 1 (64 nodes, ReLU activation function);

[0062] Hidden layer 2 (32 nodes, ReLU activation function);

[0063] Hidden layer 3 (16 nodes, ReLU activation function);

[0064] Output layer (1 node, linear activation, outputs predicted stiffness value);

[0065] Model training and optimization:

[0066] The data feature groups are randomly divided into training set, validation set and test set in a ratio of 7:2:1. The mean squared error is used as the loss function. On the training set, the backpropagation algorithm (for neural networks) or gradient boosting algorithm is used to iteratively optimize the model parameters.

[0067] After each training round, the loss is calculated on the validation set. When the validation set loss no longer decreases in multiple consecutive training rounds, the training is terminated early and rolled back to the model parameters with the lowest validation loss.

[0068] Using the validation set, key hyperparameters (such as learning rate, number of network layers and nodes, and tree depth and number of GBDTs) are tuned through grid search or Bayesian optimization to find the configuration that performs best on the validation set.

[0069] Model evaluation and determination: The model was tested using a test set, and the model with the coefficient of determination closest to 1 was selected as the solution model for the saddle-compacting condition.

[0070] Understandably, the significance of step S10 lies in constructing a "compaction model for saddle-joint compaction" using historical construction data. Specifically, this involves collecting roller sensor data, material temperature data, and actual stiffness measurements. The former two are used as input features, and the latter as the prediction target, to train a deep feedforward neural network model. This model, by learning the complex nonlinear relationship between input and output, can accurately calculate the "equivalent stiffness of new material," representing the actual compaction quality of the new material, from the mixed signal disturbed by old material. Its direct effect is to transform the stiffness measurement of the traditional CCC system, which fails in the saddle-joint area, into a reliable intelligent evaluation index that truly reflects the compaction state of the new material, laying a core data foundation for subsequent precise control.

[0071] Step S20: The joint line is identified in real time through the vision system and high-precision positioning, and the steel wheel trajectory is combined to determine whether the road roller has entered the joint compaction condition. If it has, the degree of interference of the original CCC signal is determined by analyzing the symmetry of the vibration data of the left and right steel wheels.

[0072] In step S20, the specific process for determining whether the road roller has entered the joint compaction condition is as follows:

[0073] Ground images are acquired through the airborne forward-facing camera / laser profilometer included in the vision system, and the position coordinates and attitude information (including but not limited to roll, pitch, and yaw angles) of the road roller are obtained through the GNSS / INS system on the road roller itself.

[0074] The seam line is identified by a vision system and high-precision positioning, and the spatial curve of the seam line is obtained.

[0075] The specific method for obtaining the joint line space curve can be as follows: identify the joint line in the ground image through an edge detection algorithm, convert the 2D pixel coordinates in the ground image into 3D local coordinates centered on the road roller according to the camera calibration parameters and the real-time attitude of the road roller, and combine the position coordinates and attitude of the road roller to map the joint line into a 3D space curve in the global coordinate system to obtain the joint line space curve.

[0076] Based on the roller's position coordinates and attitude information, the precise motion trajectory of the left and right steel wheel ground contact center points (also global 3D coordinates) is calculated using the dead reckoning algorithm, thus obtaining the steel wheel trajectory.

[0077] Calculate the shortest lateral distance D between the steel wheel trajectory and the spatial curve of the joint line;

[0078] If the shortest lateral distance D is less than or equal to a preset threshold (such as 1.2 times the radius of the steel wheel), it indicates that the rolling condition with the joint is entered.

[0079] If the shortest lateral distance D is greater than the preset threshold, it means that the rolling condition of saddle-to-seam has not been entered.

[0080] In step S20, the process of determining the degree of interference to the original CCC signal is as follows:

[0081] After entering the joint compaction condition, the vibration signals of the left and right steel wheels of the roller are acquired, and the ratio of the effective values ​​(RMS) of the vibration signals of the left and right steel wheels is calculated to obtain the amplitude symmetry ratio.

[0082] Compare the amplitude symmetry ratio with a preset amplitude symmetry range (e.g., 0.98-1.02);

[0083] If the amplitude symmetry ratio is within the preset amplitude symmetry range, it indicates that the original CCC signal is less affected by interference.

[0084] If the amplitude symmetry ratio is not within the preset amplitude symmetry range, it indicates that the original CCC signal is highly interfered with.

[0085] Understandably, the significance of step S20 lies in accurately identifying whether the roller has entered a joint-swapping condition and assessing the reliability of the measurement signal. The specific solution involves fusing a vision system (camera / laser) with high-precision GNSS / INS positioning. Through image recognition and coordinate transformation, a three-dimensional spatial curve of the joint line is constructed in real time, and the steel wheel trajectory is calculated. The distance between the two wheels is then used to determine if a joint has been entered. Once entered, the amplitude symmetry ratio of the vibration signals from the left and right steel wheels is calculated to quickly determine whether the original CCC signal is distorted due to differences in support on both sides. Its direct effect is to achieve automatic and accurate perception of specific compaction conditions and to complete a preliminary diagnosis of the reliability of the measurement signal, providing a clear trigger condition for deciding whether to activate the advanced solution model.

[0086] Step S30: If the degree of interference is high, the calculation model for the saddle-joint compaction condition is activated, and the old material support effect is filtered out by combining the current road construction data. The equivalent stiffness of the new material is output and compared with the target stiffness range to determine whether compaction control intervention is required.

[0087] In step S30, the process of outputting the equivalent stiffness of the new material is as follows:

[0088] The model for the rolling and compaction condition with the joint is enabled, and the current road construction data is used as the input to the model. The output is the equivalent stiffness of the new material.

[0089] The current road construction data and the historical road construction data contain the same data types;

[0090] In step S30, the process of determining whether compaction control intervention is needed is as follows:

[0091] Compare the equivalent stiffness of the new material with the target stiffness range;

[0092] If the equivalent stiffness of the new material is within the target stiffness range, then no compaction control intervention is required.

[0093] If the equivalent stiffness of the new material is less than the minimum value of the target stiffness range, then compaction control intervention is required.

[0094] It should be noted that the target stiffness range is constructed by the target stiffness and the allowable stiffness error during road construction. The reason for compaction control intervention when the equivalent stiffness of the new material is less than the minimum value of the target stiffness range is that, due to the support effect of the old material, the actual compaction requirements of the new material area are usually seriously underestimated. Therefore, the stiffness of the new material area is usually insufficient, and compaction control intervention is required.

[0095] Understandably, the significance of step S30 lies in intelligent diagnosis and decision-making when the signal is interfered with. Specifically, when S20 determines that the signal is distorted, the "straight-joint compaction condition calculation model" trained in S10 is immediately activated. Real-time construction data (vibration, temperature, velocity, etc.) are input into the model, filtering out the old material support effect and outputting a pure "new material equivalent stiffness." This stiffness value is then compared with the target stiffness range required for construction. If it falls below the lower limit, intervention is deemed necessary. The direct effect is that it achieves accurate "diagnosis" of the compaction quality of new material under complex interference environments, transforming the decision of "whether to adjust the compaction process" from relying on experience-based judgment to an objective judgment based on model calculation data, ensuring the accuracy and necessity of intervention.

[0096] Step S40: If necessary, take the equivalent stiffness of the new material as the control target, and combine it with the speed and amplitude constraints of the roller to predict and output the optimal amplitude and speed adjustment sequence.

[0097] In step S40, the process of predicting and outputting the optimal amplitude and velocity adjustment sequence is as follows:

[0098] A1. In historical road construction data, the equivalent stiffness of new material, roller speed, current amplitude level, and material temperature gradient are extracted as input features, and the equivalent stiffness of new material is used as the output feature. A state prediction model is constructed based on a deep feedforward neural network as the basic framework model.

[0099] It should be noted that the state prediction model was trained and validated using historical road construction data;

[0100] A2. Based on the roller design report, determine the amplitude constraints and speed constraints. The amplitude constraints include the available discrete speeds (such as low / medium / high), and the speed constraints include the allowable speed range (such as 2-6 km / h) and the maximum acceleration / deceleration limit.

[0101] A3. Based on the road roller control system, determine the optimization time window constraint (e.g., the optimization time for 95% control system is 3 seconds), and determine the control target requirement: at the end of the optimization time window, the equivalent stiffness of the new material reaches the minimum target stiffness or is within the target stiffness range.

[0102] A4. Divide the optimization time window into several time steps of equal length (e.g., divide 3 seconds into 10 time steps of 0.3 seconds each). Under amplitude and velocity constraints, randomly select a set of candidate amplitude-velocity operation pairs for each time step. For example, the amplitude-velocity operation pair selected for 0.9 seconds to 1.2 seconds is amplitude maintenance (medium) - velocity linearly decreasing to 3.0 km / h. Integrate all operation pairs according to the time sequence to obtain the candidate operation sequence.

[0103] A5, input the candidate operation sequence into the state prediction model, and predict the output of the new material equivalent stiffness;

[0104] A6. The first round of screening of candidate operation sequences is conducted based on the equivalent stiffness of the new material, specifically including:

[0105] Candidate operation sequences whose predicted output of the new material equivalent stiffness meets the control target requirements are retained and recorded as available operation sequences;

[0106] Candidate operation sequences whose predicted output of the new material equivalent stiffness does not meet the control target requirements will be eliminated.

[0107] A7. Perform second-order difference calculations on each available operation sequence to obtain the smoothed characterization value of the available operation sequence, and select the available operation sequence with the smallest smoothed characterization value as the optimal amplitude and velocity adjustment sequence.

[0108] For example, the optimal amplitude and velocity adjustment sequence for the predicted output is as follows:

[0109]

[0110] Understandably, the significance of step S40 lies in generating specific, optimized, and smooth operating instructions for the compaction equipment. The specific approach involves constructing a "state prediction model" and, based on the principle of Model Predictive Control (MPC), generating a large number of candidate "amplitude-velocity" operation sequences within a 3-second optimization window. The model then predicts the final compaction effect and selects the optimal sequence that achieves the required stiffness and provides the smoothest operation (evaluated by calculating the second-order difference penalty term of the velocity sequence). The direct effect is to output a specific and executable timing-based operating instruction, such as "maintain medium vibration for the first second, linearly reduce speed to 3.0 km / h, then switch to high vibration…", ensuring that compaction is increased in the most stable and efficient manner within a limited time.

[0111] Example 2

[0112] Please see Figure 3 As shown in the embodiment of the present invention, an intelligent paving and compaction control system for road construction includes the following modules:

[0113] Solution model construction module: The solution model for the rolling compaction condition with joints is constructed by training on historical road construction data;

[0114] Interference analysis and judgment module: The joint line is identified in real time through vision system and high-precision positioning, and the steel wheel trajectory is combined to determine whether the road roller has entered the joint compaction condition. If it has, the degree of interference of the original CCC signal is judged by analyzing the symmetry of the vibration data of the left and right steel wheels.

[0115] Compaction Intervention Judgment Module: If the degree of interference is high, the calculation model of the rolling condition with joints is activated, and the old material support effect is filtered out by combining the current road construction data. The equivalent stiffness of the new material is output and compared with the target stiffness range to determine whether compaction control intervention is required.

[0116] Compaction control output module: If necessary, the equivalent stiffness of the new material is used as the control target, and the optimal amplitude and speed adjustment sequence is predicted and output in combination with the speed and amplitude constraints of the roller.

[0117] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent paving and compaction control method for road construction, characterized in that: The method comprises the following steps: Step S10: Training and constructing a riding seam rolling working condition solving model through historical road construction data; The riding seam rolling working condition refers to a specific operation state of a steel wheel of a road roller transversely crossing an old road surface that has been compacted and cooled and new hot soft material that is newly paved during longitudinal joint construction of an asphalt pavement; During the training and construction of the riding seam rolling working condition solving model, a gradient boosting decision tree or a deep feedforward neural network is selected as a basic model architecture, the road roller body sensing data and the material temperature data are selected as input features, and the real stiffness data is selected as output features to integrate into a data feature group, and all data in the data feature group are collected at the same time and from the same new and old material area; During the training and construction of the riding seam rolling working condition solving model, the data feature group is divided into a training set, a validation set and a test set, a mean square error is used as a loss function, and a back propagation algorithm or a gradient boosting algorithm is used to iteratively optimize model parameters on the training set; After each round of training, the loss on the validation set is calculated, and when the validation set loss does not decrease for a plurality of consecutive training rounds, the training is terminated in advance, and the model parameters with the lowest validation loss are rolled back; The learning rate, the number of network layers and nodes, the tree depth and number of GBDT, and other key hyperparameters are optimized through grid search or Bayesian optimization using the validation set, and finally the test set is used for model testing, and the model with the coefficient of determination closest to 1 is selected as the riding seam rolling working condition solving model; Step S20: Real-time identification of the joint line through a visual system and high-precision positioning, and judgment of whether the road roller enters the riding seam rolling working condition in combination with the steel wheel track, if yes, judgment of the degree of interference of the original CCC signal by analyzing the symmetry of the left and right steel wheel vibration data; Step S30: If the degree of interference is high, the riding seam rolling working condition solving model is enabled, the old material support effect is filtered out in combination with the current road construction data, the new material equivalent stiffness is output, and comparison is made with the target stiffness range to determine whether compacting control intervention is needed; Step S40: If needed, the new material equivalent stiffness is taken as a control target, a combination of the road roller speed and the amplitude constraint condition is used to predict and output an optimal amplitude and speed adjustment sequence; The process of predicting and outputting the optimal amplitude and speed adjustment sequence comprises: Dividing the optimization time window into a plurality of equal-length time steps, randomly selecting a group of candidate amplitude-speed operation pairs for each time step under the amplitude constraint and the speed constraint, and integrating all operation pairs in sequence to obtain a candidate operation sequence; The candidate operation sequence is input into a state prediction model to predict and output the new material equivalent stiffness, and the candidate operation sequence that meets the control target requirement is selected as an available operation sequence, and the candidate operation sequence that does not meet the control target requirement is filtered out; Second-order difference calculation is performed on each available operation sequence to obtain a smooth representation value of the available operation sequence, and the available operation sequence with the smallest smooth representation value is selected as the optimal amplitude and speed adjustment sequence. The construction of the state prediction model is: extracting the new material equivalent stiffness, the speed of the road roller, the current amplitude range, and the material temperature gradient from the historical road construction data as input features, taking the new material equivalent stiffness as an output feature, using a deep feedforward neural network as a basic framework model, and training and verifying the state prediction model through historical road construction data; The constraint conditions include amplitude constraints and speed constraints, the amplitude constraints include available discrete ranges, the speed constraints include allowed speed ranges and maximum acceleration and deceleration limits, and the optimization time window constraint is determined according to the road roller control system, and the control target requires that the new material equivalent stiffness reaches the minimum target stiffness or is within the target stiffness range at the end of the optimization time window.

2. The intelligent paving and compaction control method for road construction according to claim 1, characterized in that: The historical road construction data includes: road roller body sensing data, specifically original CCC signals, high-frequency acceleration time domain data of left and right steel wheel independent vibration, road roller amplitude, frequency and driving speed; Material temperature data, including the coverage ratio of the new material area in the joint area, the average temperature of the new material area, the average temperature of the old material area, and the temperature gradient of the new and old material areas; Real stiffness data, specifically the field detection value of the joint new material side, obtained by point detection with a portable falling weight deflectometer and a nuclear density meter after joint rolling.

3. The intelligent paving and compaction control method for road construction according to claim 1, characterized in that: The specific operation of determining whether the road roller enters the joint rolling working condition is as follows: Collect ground images through a vision system, and obtain road roller position coordinates and attitude information through a road roller body GNSS / INS system; Convert 2D pixel coordinates in the ground image into 3D local coordinates centered on the road roller according to camera calibration parameters and real-time attitude of the road roller, map the joint line into a 3D space curve in the global coordinate system by combining the road roller position coordinates and attitude, and obtain the joint line space curve; According to the road roller position coordinates and attitude information, the accurate motion trajectories of the left and right steel wheel contact center points are calculated by using the dead reckoning algorithm, the shortest transverse distance between the steel wheel trajectory and the joint line space curve is calculated, and if the distance meets the requirements of entering the joint rolling working condition, it is determined that the joint rolling working condition is entered, otherwise it is not entered.

4. The intelligent paving and compaction control method for road construction according to claim 1, characterized in that: The operation of determining the interference degree of the original CCC signal is as follows: After confirming that the joint rolling working condition is entered, the left and right steel wheel vibration signals of the road roller are obtained, the amplitude symmetry ratio of the effective values of the left and right steel wheel vibration signals is calculated, and the amplitude symmetry ratio is obtained; If the amplitude symmetry ratio is not within the amplitude symmetry range, it indicates that the original CCC signal has a high interference degree, otherwise, the interference degree is low.

5. The intelligent paving and compaction control method for road construction according to claim 1, characterized in that: When outputting the new material equivalent stiffness, the constructed joint rolling working condition calculation model is enabled, the current road construction data of the same type as the historical road construction data is input, the old material support effect is filtered out through model operation, and the new material equivalent stiffness is directly output; When judging whether to need to perform the compaction control intervention, the new material equivalent stiffness is compared with the target stiffness range, the target stiffness range is constructed by the target stiffness and the stiffness allowable error during road construction, and if the new material equivalent stiffness is within the target stiffness range, the compaction control intervention is not needed. If the new material equivalent stiffness is less than the minimum value of the target stiffness range, the old material support effect usually seriously underestimates the actual compaction demand of the new material region, the stiffness of the new material region is insufficient, and therefore the compaction control intervention is needed.

6. An intelligent paving and compaction control system for road construction, characterized in that, The system is used for executing the method in any one of claims 1-5, and the system comprises: a calculation model construction module that constructs a joint rolling working condition calculation model through historical road construction data; an interference analysis and judgment module that identifies a joint line in real time through a visual system and high-precision positioning, judges whether a road roller enters a joint rolling working condition in combination with a steel wheel track, if entering, judges the interference degree of an original CCC signal by analyzing the symmetry of left and right steel wheel vibration data; a compaction intervention judgment module that, if the interference degree is high, enables the joint rolling working condition calculation model, filters out the old material support effect in combination with current road construction data, outputs the new material equivalent stiffness, and compares the new material equivalent stiffness with a target stiffness range to judge whether to need to perform the compaction control intervention; a compaction control output module that, if needed, takes the new material equivalent stiffness as a control target, combines a road roller speed and amplitude constraint condition, and predicts and outputs an optimal amplitude and speed adjustment sequence.

Images